[PATCH] D134064: [InstCombine] Baseline tests for folding ((x?1:4)&(y?1:4))==0 to x^y

[Sanjay Patel via Phabricator via llvm-commits]

spatel added inline comments.


================
Comment at: llvm/test/Transforms/InstCombine/select_and_icmpeq.ll:151
+;
+  %z = add i6 1, 2
+  %s1 = select i1 %x, i6 %z, i6 5
----------------
marcauberer wrote:
> spatel wrote:
> > This (and similar other tests) doesn't provide the coverage that you were hoping for. 
> > This will be constant-folded before we ever reach the transform that we're hoping to unit-test.
> > That's why the test is already completely reduced to "false". :)
> Okay, shall I remove them?
Yes - we don't want tests for other transforms as part of this patch.
But we still need tests to verify that the transform to be added is behaving as expected.
See if something like this works:
https://alive2.llvm.org/ce/z/2er3z7
...and adjust as needed (make the false arm of the selects non-immediate too), so the tests provide coverage for more of the pattern variations.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D134064/new/

https://reviews.llvm.org/D134064